In general, to generate a composite image, a plurality of partially overlapping images is taken by rotating an image capturing device along certain axes. These separate images are later placed together to form a composite image. Even though overlapping regions are carefully placed together, the composite image often lacks the naturally continuous appearance. To illustrate the above problem, referring to FIG. 1A, a digital camera d is positioned at a predetermined location and is rotated along a first axis which is parallel to a z axis so as to take images a and b. Then, still at the same location, the digital camera d is now rotated along a x axis to take an image c. The corresponding filming ranges or areas are shown in the respective boundaries. The images a, b and c are projected onto a corresponding image plane a', b' or c'. Referring to FIG. 1B, when the above described images on the image plane a', b' and c' are placed together, certain portions of the images are distorted from the original image as shown in FIG. 1A.
Referring to FIG. 2, the above described problem is caused by the following reasons. An image capturing device such as a digital camera d captures an image of an object 3. The camera d is rotated to a position d' along a common axis. As a result of rotation, a lens at a first position 1 is moved to a second position 1' while an image capturing plane such as a charge coupled device (CCD) is rotated to a position 2'. A portion bc of the object 3 is captured as an image PlP2 on the CCD 2. After the camera is rotated to the position d', the same image portion bc is now captured as an image qr on the CCD at the position 2'. Even if no distortion is assumed for the lens, because of the different angles of the CCD 2 with respect to the object 3, the same image portion bc is formed with a different length.
Composite pictures are often created when an image is too large for a single image-carrying medium. For example, a panoramic view of astronomical images is generated from a series of individual images covering a partial view. Since a relative angle of each of the individual images can be determined, it is possible to correct the individual images based upon the relative angle. However, it is not always the case that the relative angle of the individual image is available.